Keerthi Vasan M

Dual Degree Student(B.Tech+M.Tech) · Indian Institute of Technology Madras ·

  • I am an Dual Degree student in my 9th semester pursuing BTech in Engineering Design and MTech in Robotics at Indian Institute of Technology Madras.
  • Interned at Ola Krutrim (India’s leading AI startup) and ISRO (India’s flagship space research agency), gaining hands-on experience in robotics and advanced technologies.
  • I am a research enthusiastic person who aims to develop cutting edge robotic technologies.
  • I was previously a part of Team Abhiyaan where i was involved in the development of autonomous golf cart and AMRs for International Ground Vehicle Competition which takes place in United States every year
  • I am passionate about exploring how robot learning can drive adaptive and efficient interactions in real-world environments. Specifically, I study intelligent manipulation and human–robot collaboration, aiming to create systems that learn, adapt, and work alongside people naturally.
  • In my free time I enjoy listening to music, playing football, singing and painting
  • My first-author paper was recently accepted at ICRA 2025, Atlanta — one of the flagship international conferences in robotics.


Download CV / Resume

You can download my CV here (Last updated March, 2024)

Or view the embedded PDF below:




Skills

Programming: ROS , ROS2 , Linux , OpenCv , Pytorch , C , C++ , python , Java , Matlab

Artificial Intelligence: Reinforcement Learning , CNN ,Deep Learning

Simulators: Gazebo , Webots

Works/ Projects

Trajectory Tracking Pipeline for a Holonomic System

This project is centered on following a specified trajectory in the task space of varying shapes using the end effector of the Kuka Youbot, while navigating around obstacles. We introduce a novel inverse kinematics solution method to convert task space coordinates into configuration space coordinates. By employing RRT* or BiRRT* in the 8-degree-of-freedom (DOF) system, we plan the trajectory efficiently. The outcomes illustrate that our algorithm adeptly tracks the pattern utilizing near-optimal sampling-based algorithms like RRT* and BiRRT*, both in the presence and absence of obstacles. Our methodology brings notable advancements in inverse kinematics determination for 8 DOF systems, transforming the conventional control tracking challenge into a motion planning problem formulation that remains effective even in obstacle-laden scenarios, unlike traditional control approaches.

Identifying Lanes and getting lane Polynomials for better decision making for autonomous vehicle

This project is aimed at identifying and converting lane mask data, obtained from the YOLOP algorithm, into quadratic polynomials in the world frame using various algorithms such as the Pinhole Camera Model and DBSCAN. The clustering of lanes is highly effective, allowing for accurate polynomial extraction. The basic tracking algorithm operates on the Markov first-order principle, offering slight improvements in lane tracking.

The algorithm performs exceptionally well when provided with high-quality lane data, as demonstrated in the following video:

In this test, the algorithm initially fails due to poor lane data but demonstrates high performance once accurate lane data is obtained.

Advanced Driver Assistant System for Heavy Vehicles

Truck drivers are at fault in 90% of two-vehicle crashes involving trucks, according to a report published by Volvo in 2017 by its accident research team. Over 4000 people lose their life to such heavy vehicle related accidents. If we were able to assist the driver, we could save about 3600 people. Most of the Blindspot region is concentrated to the left side of the truck relative to the direction of motion for a right hand drive truck.

Our aim is to create a Driver Assistance System, particularly for Heavy Vehicles using an array of Positional Sensors around the trailer to perceive and track objects in the blind spots and relay this information to the driver through haptic feedback. The system will help to augment the driver’s perception of his surroundings, providing additional redundancy in case the driver loses awareness of a part of their vehicle’s surroundings. Feedback will only be provided when the driver may potentially be colliding with an obstacle, such as when there’s a car and the driver tries to turn.

Inch Worm Based Robotic Arm

This project unveils a dynamic robotic arm designed to navigate through space utilizing inverse kinematics planning. Departing from the conventional fixed-base model, this arm showcases unparalleled versatility by maneuvering within the spatial environment. Its revolutionary design serves as a pivotal solution, particularly suited for space station operations where agile and adaptable movement capabilities are imperative. Developed and implemented through the Robot Operating System (ROS) and tested in gazebo, this advanced arm technology marks a significant leap forward in the realm of space robotics, offering unprecedented flexibility and precision in spatial exploration and manipulation tasks